setup-and-data
Set up NAVSIM v2 dependencies and workspace data, select a supported split, validate paths, and use the Scene, Frame, AgentInput, SensorConfig, and SceneLoader contracts.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- setup-and-data
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Set up NAVSIM v2 dependencies and workspace data, select a supported split, validate paths, and use the Scene, Frame, AgentInput, SensorConfig, and SceneLoader contracts.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/data-api.md5.8 KB
- 参考文档references/data-layout.md6.1 KB
- 参考文档references/installation.md3.7 KB
- 参考文档references/troubleshooting.md6.7 KB
03 SKILL.md 正文
NAVSIM setup and data
Use this route before training, evaluation, or submission work. It is a read-only setup and data contract: it may inspect an existing environment and workspace, but it must not download archives, delete partial downloads, create large caches, train, score, or upload anything by default.
Route
-
Read installation.md for the Python/dependency baseline and environment variables.
-
Read data-layout.md to map a selected split to logs, original sensors, synthetic scenes/sensors, maps, and experiment/cache roots. Treat original and synthetic sensor/log/cache roots as different resources.
-
Run the bundled, side-effect-free workspace validator before a data-dependent runner:
python scripts/validate_workspace.py --split miniA successful run ends with
VALIDATION PASSED; failures identify the missing variable or path and exit nonzero. Use--helpfor split names and--jsonfor machine-readable diagnostics. The validator does not read a project config, download data, or create directories. -
Read data-api.md when constructing a loader or agent input. Use no-sensor loading first when checking metadata/filtering; enable only the camera history and/or LiDAR iterations actually required by the agent.
-
For failures, use troubleshooting.md and preserve the distinction between an unavailable optional backend and a broken required data contract.
Split guardrails
mini,trainval, andtestare OpenScene log/sensor splits.navminiandnavtrainare filtered NAVSIM views overminiandtrainval;navtestand the v2 two-stage test views are filtered views overtest. A filtered view does not imply a separate complete log root.navhard_two_stage,warmup_two_stage, andprivate_test_hard_two_stagerequire synthetic/competition assets in addition to the appropriate original assets. Select their matching bundle; do not silently reusenavhard_two_stageassets for warmup or private data.- Never use
test,navtest,navhard_two_stage,warmup_two_stage, orprivate_test_hard_two_stageas training data for a competition submission. See the split matrix in data-layout.md.
The nearest references are the authoritative distilled operating context for this sub-skill. They intentionally describe interfaces and layouts rather than asking a later agent to open or execute source-checkout files.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.